39Hydro Merging and Downscaling of Satellite and Model Soil Moisture over Taiwan

Soil moisture (SM) plays a crucial role in modulating the hydroclimate system, yet obtaining an accurate, high-resolution estimate of SM is challenging. Various techniques for obtaining SM data include satellite sensors, land surface models, and in-situ measurements; however, no single techniques exhibit an absolute advantage over the others in terms of accuracy and spatiotemporal coverage. This study employs the Triple Collocation (TC) method to merge SM data from the Climate Change Initiative (CCI), Soil Moisture Active Passive (SMAP), and the High-Resolution Land Data Assimilation System (HRLDAS) products over Taiwan. The TC method involves random error analysis used to calculate the weight of each SM product for the weighted average-based merging of SM. Afterwards, the merged SM product at the grid spacing of 0.25° is downscaled to 3 km based on a deep learning method referred to as Convolutional Long Short-Term Memory Networks (ConvLSTM), with the use of SM-related environmental factors (e.g., topographic attributes, soil properties, land use, surface reflectance, temperature, solar radiation, precipitation, and vegetation indices) as input variables. All the original, by-products (merged or downscaled), and final (merged and downscaled) products are validated against in-situ measurements. The merged and downscaled SM that shows the overall highest performance and resolves the data scarcity issue is expected to facilitate various hydrometeorological, agricultural, and water resources applications.